handbook on impact evaluation conclusions

《handbook on impact evaluation conclusions》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。

1、10.ConclusionsImpactevaluationmethodsexaminewhetherprogrameffectscanbeidentified.Thatis,theyseektounderstandwhetherchangesinsuchoutcomesasconsumptionandhealthcanbeattributedtotheprogramitselfandnottosomeothercause.Thishandbookdescribesmajorquantitativemetho

2、dsthatareprimarilyusedinexpostimpactevalu-ationsofprogramsandpolicies.Italsodiscusseshowdistributionalimpactscanbemeasured,aswellasexanteapproachesthatpredicttheoutcomesofprogramsandmechanismsbywhichprogramsaffecttargetedareas.Randomizedevaluationsseektoid

3、entifyaprogramseffectbyidentifyingagroupofsubjectssharingsimilarobservedcharacteristics(say,acrossincomesandearningopportunities)andassigningthetreatmentrandomlytoasubsetofthisgroup.Thenontreatedsubjectsthenactasacomparisongrouptomimiccounterfactualoutcome

4、s.Thismethodavoidstheproblemofselectionbiasfromunobservedcharacteristics.Randomizedevaluations,however,maynotalwaysbefeasible.Insuchcases,research-ersthenturntoso-callednonexperimentalmethods.Thebasicproblemwithanon-experimentaldesignisthatforthemostpartin

5、dividualsarenotrandomlyassignedtoprograms,andasaresultselectionbiasoccursinassessingtheprogramimpact.Thisbookdiscussesanumberofapproachesthataddressthisproblem.Propensityscorematchingmethods,forexample,attempttoreducebiasbymatchingtreatmentandcon-trolhouse

6、holdsonthebasisofobservablecovariates.Propensityscorematchingmeth-odsthereforeassumethatselectionbiasisbasedonlyonobservedcharacteristicsandcannotaccountforunobservedheterogeneityinparticipation.Morecanbelearnedifoutcomesaretrackedforbothparticipantsandnon

7、partici-pantsoveratimeperiodthatisdeemedsufficienttocaptureanyimpactsoftheinter-vention.Apopularapproachinnonexperimentalevaluationisthedouble-difference(ordifference-in-difference)method,althoughthismethodcanalsobeusedinexperi-mentalapproaches.Thismethodpo

8、stulatesthatifoutcomesforbothparticipantsandnonparticipantsaretrackedoveraperiodoftime,suchtrackingwouldprovideagoodbasisforidentifyingtheprogrameffect.Sowithdoubledifference,theobservedchangesovertimefornonp